Lacrosse Analytics

D1 Mens Lacrosse
User avatar
CU77
Posts: 3474
Joined: Thu Jul 26, 2018 1:49 pm

Re: Lacrosse Analytics

Post by CU77 »

Link doesn't work ...
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

laxreference wrote: Tue May 07, 2019 3:40 pm If you are not loving the reliance on RPI, you might like SOR better.

CU77 wrote: Tue May 07, 2019 5:01 pm Link doesn't work ...
Oops. It's been fixed. Thanks.
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
User avatar
CU77
Posts: 3474
Joined: Thu Jul 26, 2018 1:49 pm

Re: Lacrosse Analytics

Post by CU77 »

you take a generic “good” team (I use the #10 Lax-ELO team), and you estimate how many wins they would end up with against that schedule.
And how do you do this estimate? Lax-ELO?

Seems to me the whole idea is entirely dependent on the underlying rating system that's used to estimate win probabilities. And if so, why not just sort according to that rating?
User avatar
Hawkeye
Posts: 528
Joined: Sun Dec 09, 2018 4:51 pm

Re: Lacrosse Analytics

Post by Hawkeye »

CU77 wrote: Tue May 07, 2019 7:26 pm
you take a generic “good” team (I use the #10 Lax-ELO team), and you estimate how many wins they would end up with against that schedule.
And how do you do this estimate? Lax-ELO?

Seems to me the whole idea is entirely dependent on the underlying rating system that's used to estimate win probabilities. And if so, why not just sort according to that rating?
These are very good questions.
Georgia Tech alumnus
2019 Georgia Tech lacrosse final record: 18-2; MCLA semifinalist
-
College lacrosse television schedules: https://markmhart9.wixsite.com/mysite
Big Dog
Posts: 490
Joined: Fri Mar 15, 2019 8:18 pm

Re: Lacrosse Analytics

Post by Big Dog »

besides HP, the other team that jumps out in the SOR is the high ranking of tOSU, who couldn't even qualify for thier own league tourney. (5th man out based on tie breakers.)
User avatar
Hawkeye
Posts: 528
Joined: Sun Dec 09, 2018 4:51 pm

Re: Lacrosse Analytics

Post by Hawkeye »

Big Dog wrote: Tue May 07, 2019 7:43 pm besides HP, the other team that jumps out in the SOR is the high ranking of tOSU, who couldn't even qualify for thier own league tourney. (5th man out based on tie breakers.)
Ohio State actually finished last in the Big Ten this season. The 2-3 tiebreaker didn't play out.

1. PSU 5-0
2. JHU 3-2 (head to head with UMCP)
3. UMCP 3-2
4. RU 2-3
5. UM 1-4 (head to head with tOSU)
6. tOSU 1-4
Georgia Tech alumnus
2019 Georgia Tech lacrosse final record: 18-2; MCLA semifinalist
-
College lacrosse television schedules: https://markmhart9.wixsite.com/mysite
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

Hawkeye wrote: Tue May 07, 2019 7:50 pm
Big Dog wrote: Tue May 07, 2019 7:43 pm besides HP, the other team that jumps out in the SOR is the high ranking of tOSU, who couldn't even qualify for thier own league tourney. (5th man out based on tie breakers.)
Ohio State actually finished last in the Big Ten this season. The 2-3 tiebreaker didn't play out.

1. PSU 5-0
2. JHU 3-2 (head to head with UMCP)
3. UMCP 3-2
4. RU 2-3
5. UM 1-4 (head to head with tOSU)
6. tOSU 1-4
OSU also doesn't really stand out in SOR. They were 14th in the final RPI and they have the 14th strongest resume according to SOR.
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

CU77 wrote: Tue May 07, 2019 7:26 pm
you take a generic “good” team (I use the #10 Lax-ELO team), and you estimate how many wins they would end up with against that schedule.
And how do you do this estimate? Lax-ELO?

Seems to me the whole idea is entirely dependent on the underlying rating system that's used to estimate win probabilities. And if so, why not just sort according to that rating?
Yep, in my implementation, the excess wins you earn from a given win would be determined by the pre-game win probability, which comes from Lax-ELO. You could use any model for predicting the pre-game win probabilities though. ELO is just the only one that I've got for doing that specific step. I do think it's important to separate the SOR approach (which could leverage any pre-game win probability calculation) from the way that I do it. Would hate for people to discredit SOR simply because they don't agree with using ELO ratings to predict outcomes.

As far as using ELO instead of SOR, I suppose you could do that, but since ELO systems do not start each team at the same level going into a new season, your final selections would explicitly include the previous year's results. Using ELO to come up with the pre-game win probabilities does this too, but much more indirectly.
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
User avatar
DALaxDad
Posts: 295
Joined: Thu Aug 02, 2018 12:52 pm

Re: Lacrosse Analytics

Post by DALaxDad »

Someone asked about impact of rule changes. I posted this elsewhere for what it is worth:

Saves per game for the 15 goaltenders with the highest number of saves per game average the following: for 2019 (through 4/24) 13.58 with Edelmann at #1 with 15.09 and Boyce at #15 with 12.64; and for 2018 (season), 12.07 with Braun at #1 with 13.53 and Heger at #15 with 11.31.

Average saves per game of the top 15 are up by 1.51 saves per game.
User avatar
CU77
Posts: 3474
Joined: Thu Jul 26, 2018 1:49 pm

Re: Lacrosse Analytics

Post by CU77 »

laxreference wrote: Tue May 07, 2019 8:25 pmAs far as using ELO instead of SOR, I suppose you could do that, but since ELO systems do not start each team at the same level going into a new season, your final selections would explicitly include the previous year's results. Using ELO to come up with the pre-game win probabilities does this too, but much more indirectly.
Indirectly or not, last year's results influence the final SOR if the win-probability system uses last year's results. (Kinda obvious, no?) But you could do Lax-ELO with a cold start (all teams start the season with the same rating).

More to the point, we don't need multiple metrics. At the end of Selection Sunday, we need an ordered list of teams. That's all anyone cares about, really. If we could all agree on a formula for this (as has been done in hockey), the sturm-und-drang of committee selection could be eliminated.
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

And deprive those fine folks of their weekend in the hotel? The outrage!

I would love to have an agreed upon formula so that there is no subjectivity. It worked for the BCS except that they made the cut-off too high (2 vs 3) and they used human polls.

No one is going to complain if the 13th team on the list narrowly misses the field, especially because if it's an agreed formula, we can project where teams will fall depending on the outcomes of games yet to be played. Teams would more or less know ahead of time what they need to do or whether they are at risk of bid-thieves.

I'll sign your petition
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
User avatar
CU77
Posts: 3474
Joined: Thu Jul 26, 2018 1:49 pm

Re: Lacrosse Analytics

Post by CU77 »

laxreference wrote: Wed May 08, 2019 2:15 pmAnd deprive those fine folks of their weekend in the hotel? The outrage!
The hockey committee still gets their weekend. They actually have full discretion, but have tacitly agreed to follow the formula for selection of the at-large teams, and have done so for the past 20 or so years. They have some discretion for seeding, which they use to eliminate first-round match-ups of teams from the same conference, and to try to put teams in a nearby regional (the first two rounds of the 16-team tournament are at pre-selected regional sites), but still follow the formula as closely as possible.

One year they did not do this, swapping the top-2 teams because #2 (by the formula) had beaten #1 (by the formula) in a conference-tournament final. Outrage followed immediately in all quarters: they didn't follow the formula!!! The committee learned its lesson and stuck to their knitting in subsequent years.

The formula is essentially RPI, but with fractions of 25% for own record, 22% for opponents' record, and 53% for opponents-opponents record, and different weights for home and away games. Also, if your RPI drops because you beat a weak team, that game is dropped for you: there is no penalty for winning (unlike lacrosse). The percentages are a bit weird. They actually have the effect of enhancing the importance of your own record, since opponents-opponents is very tightly bunched around 0.500 for all teams. A team like High Point in lacrosse would do well by this formula, a team like JHU would not.
laxreference wrote: Wed May 08, 2019 2:15 pmNo one is going to complain if the 13th team on the list narrowly misses the field, especially because if it's an agreed formula, we can project where teams will fall depending on the outcomes of games yet to be played. Teams would more or less know ahead of time what they need to do or whether they are at risk of bid-thieves.
Exactly right. This year Minnesota missed out on the tournament by a thousandth of a point on the formula (I think it was Minnesota: I haven't gone back to check). The team's fans shrugged their shoulders (on the equivalent of this forum) and said oh well, we hope to do better next year. And, as you say, fan sites track the formula and show who will be in or out depending on how the conference tournaments go. It's actually a pretty exciting way to do things.

As long as the rules are obviously fair and unbiased, everyone accepts them.
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

Interesting that they switched the weights in that way. But I guess it makes sense. They used the adjustment to dampen the impact of the 2nd criteria rather than necessarily raise the impact of your own record.

I just ran my RPI calculator with those weights instead of the current weightings (current actual RPI is here):

1) Penn State 0.657123670439
2) Virginia 0.619125342911
3) Penn 0.612794193746
4) Yale 0.608504406913
5) Duke 0.606737730048
6) Maryland 0.604966178956
7) Loyola MD 0.59635436801
8) Towson 0.594134137644
9) Syracuse 0.593969625504
10) Georgetown 0.577131946126
11) Cornell 0.57683104099
12) Army 0.576336240619
13) Notre Dame 0.576292528511
14) High Point 0.575900147304
15) Ohio State 0.574689222086
16) Denver 0.571521964969
17) Johns Hopkins 0.570778130542
18) North Carolina 0.549340947066
19) Massachusetts 0.547177741584
20) Richmond 0.544823193709
21) Villanova 0.540597044712
22) Delaware 0.539600629928
23) Air Force 0.537663369347
24) Boston U 0.534033673784
25) Hobart and William 0.530198528587
26) Rutgers 0.526268258151
27) Princeton 0.523932091936
28) Lehigh 0.521191459136
29) Marist 0.517717560438
30) Robert Morris 0.511667089035
31) Mount St Marys 0.510831368345
32) Drexel 0.508095090117
33) Stony Brook 0.506943274735
34) Navy 0.505816585021
35) Holy Cross 0.503273863454
36) Cleveland State 0.503070045662
37) Brown 0.502786664681
38) Sacred Heart 0.501983088496
39) Marquette 0.49725370923
40) Bucknell 0.491190907377
41) Vermont 0.489348894531
42) Michigan 0.487051788082
43) Jacksonville 0.486138824282
44) Providence 0.484324321215
45) Detroit 0.482551926105
46) Harvard 0.481259517124
47) Quinnipiac 0.481042265123
48) Saint Joseph's 0.477567846842
49) Hofstra 0.474379297972
50) Canisius 0.47088308036
51) Albany 0.465908378133
52) UMBC 0.461565004317
53) Colgate 0.46001526408
54) Fairfield 0.458296566987
55) Utah 0.453710124369
56) Siena 0.449629807222
57) St. John's 0.435210563472
58) Furman 0.427400577967
59) Massachusetts-Lowell 0.421978452405
60) Lafayette 0.420320432766
61) Manhattan 0.416386902304
62) Monmouth 0.41346410429
63) Bellarmine 0.406584916595
64) VMI 0.40524180939
65) Bryant 0.403825675057
66) Hartford 0.400886298577
67) Dartmouth 0.393192028446
68) Binghamton 0.383845563436
69) Mercer 0.376650246431
70) NJIT 0.358688446372
71) Wagner 0.357039649621
72) St. Bonaventure 0.356163218228
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
reLAX
Posts: 87
Joined: Wed Jan 30, 2019 9:34 am

Re: Lacrosse Analytics

Post by reLAX »

laxreference wrote: Wed May 08, 2019 3:47 pm Interesting that they switched the weights in that way. But I guess it makes sense. They used the adjustment to dampen the impact of the 2nd criteria rather than necessarily raise the impact of your own record.

I just ran my RPI calculator with those weights instead of the current weightings (current actual RPI is here):

1) Penn State 0.657123670439
2) Virginia 0.619125342911
3) Penn 0.612794193746
4) Yale 0.608504406913
5) Duke 0.606737730048
6) Maryland 0.604966178956
7) Loyola MD 0.59635436801
8) Towson 0.594134137644
9) Syracuse 0.593969625504
10) Georgetown 0.577131946126
11) Cornell 0.57683104099
12) Army 0.576336240619
13) Notre Dame 0.576292528511
14) High Point 0.575900147304
15) Ohio State 0.574689222086
16) Denver 0.571521964969
17) Johns Hopkins 0.570778130542
18) North Carolina 0.549340947066
19) Massachusetts 0.547177741584
20) Richmond 0.544823193709
21) Villanova 0.540597044712
22) Delaware 0.539600629928
23) Air Force 0.537663369347
24) Boston U 0.534033673784
25) Hobart and William 0.530198528587
26) Rutgers 0.526268258151
27) Princeton 0.523932091936
28) Lehigh 0.521191459136
29) Marist 0.517717560438
30) Robert Morris 0.511667089035
31) Mount St Marys 0.510831368345
32) Drexel 0.508095090117
33) Stony Brook 0.506943274735
34) Navy 0.505816585021
35) Holy Cross 0.503273863454
36) Cleveland State 0.503070045662
37) Brown 0.502786664681
38) Sacred Heart 0.501983088496
39) Marquette 0.49725370923
40) Bucknell 0.491190907377
41) Vermont 0.489348894531
42) Michigan 0.487051788082
43) Jacksonville 0.486138824282
44) Providence 0.484324321215
45) Detroit 0.482551926105
46) Harvard 0.481259517124
47) Quinnipiac 0.481042265123
48) Saint Joseph's 0.477567846842
49) Hofstra 0.474379297972
50) Canisius 0.47088308036
51) Albany 0.465908378133
52) UMBC 0.461565004317
53) Colgate 0.46001526408
54) Fairfield 0.458296566987
55) Utah 0.453710124369
56) Siena 0.449629807222
57) St. John's 0.435210563472
58) Furman 0.427400577967
59) Massachusetts-Lowell 0.421978452405
60) Lafayette 0.420320432766
61) Manhattan 0.416386902304
62) Monmouth 0.41346410429
63) Bellarmine 0.406584916595
64) VMI 0.40524180939
65) Bryant 0.403825675057
66) Hartford 0.400886298577
67) Dartmouth 0.393192028446
68) Binghamton 0.383845563436
69) Mercer 0.376650246431
70) NJIT 0.358688446372
71) Wagner 0.357039649621
72) St. Bonaventure 0.356163218228

Which of the two calculations are more objective? And perhaps a fairer assessment?
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

Live Win Probability Links (NCAA Tournament Edition)

- Marist vs UMBC
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

reLAX wrote: Wed May 08, 2019 4:14 pm
laxreference wrote: Wed May 08, 2019 3:47 pm Interesting that they switched the weights in that way. But I guess it makes sense. They used the adjustment to dampen the impact of the 2nd criteria rather than necessarily raise the impact of your own record.

I just ran my RPI calculator with those weights instead of the current weightings (current actual RPI is here):

1) Penn State 0.657123670439
2) Virginia 0.619125342911
3) Penn 0.612794193746
4) Yale 0.608504406913
5) Duke 0.606737730048
...

Which of the two calculations are more objective? And perhaps a fairer assessment?
They are both equally objective. Using the hockey weights reduces the impact of your strength of schedule relative to your record. So if you think RPI weights a team's schedule too much, you might prefer the hockey method.

But it's the same exact approach, it only differs on how much they weight your record vs your opponents' strength. So they are equally objective.

As to fair, I think that is in the eye of the beholder rather than an characteristic of either calculation.
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
User avatar
CU77
Posts: 3474
Joined: Thu Jul 26, 2018 1:49 pm

Re: Lacrosse Analytics

Post by CU77 »

Agreed. They're both equally "objective", and which is "fairer" depends on your point of view. The hockey formula likes a good record against a weak schedule more than a weak record against a strong schedule. So: High Point is RPI #14 (hockey) vs #20 (lax), whereas JHU is #17 (hockey) vs #8 (lax).

My personal favorite methodology is Zermelo/Bradley-Terry/KRACH, which has JHU #12 and High Point #14, and is the go-to choice of statisticians if the only info you're going to use is wins and losses (and not scores and not locations [home field advantage] and not timing [early-season vs late-season] and not last season's results):

https://mattcarberry.com/ZRatings/Z-MLXS.HTM

BUT: teams with weak records against strong schedules are often actually very good. Poster child for this is UNC 2016, which was 8-6 and KRACH #15 at the end of the regular season, and would have been left out of the tournament that they went on to win if KRACH ratings were used for selection (but I favor them anyway, no system is perfect).
reLAX
Posts: 87
Joined: Wed Jan 30, 2019 9:34 am

Re: Lacrosse Analytics

Post by reLAX »

laxreference wrote: Wed May 08, 2019 7:52 pm
reLAX wrote: Wed May 08, 2019 4:14 pm
laxreference wrote: Wed May 08, 2019 3:47 pm Interesting that they switched the weights in that way. But I guess it makes sense. They used the adjustment to dampen the impact of the 2nd criteria rather than necessarily raise the impact of your own record.

I just ran my RPI calculator with those weights instead of the current weightings (current actual RPI is here):

1) Penn State 0.657123670439
2) Virginia 0.619125342911
3) Penn 0.612794193746
4) Yale 0.608504406913
5) Duke 0.606737730048
...

Which of the two calculations are more objective? And perhaps a fairer assessment?
They are both equally objective. Using the hockey weights reduces the impact of your strength of schedule relative to your record. So if you think RPI weights a team's schedule too much, you might prefer the hockey method.

But it's the same exact approach, it only differs on how much they weight your record vs your opponents' strength. So they are equally objective.

As to fair, I think that is in the eye of the beholder rather than an characteristic of either calculation.
Thank you for that explanation.
laxreference
Posts: 1099
Joined: Wed Aug 29, 2018 3:54 pm
Contact:

Re: Lacrosse Analytics

Post by laxreference »

Let's assume that every team wins their opening round game. If you look at their projected title odds in that scenario and subtract their current odds, you get a list of the teams for whom a win would have the largest effect in terms of changing our perception of them. Even if Penn State wins their game, it's not really going to change our perception of their chances much. Yale vs Georgetown on the other hand: different story.

Here is the full post
Data Engineer/Lacrosse Fan --- Twitter: @laxreference --- Informed fans get Expected Goals, the new daily newsletter from LacrosseReference
User avatar
Hawkeye
Posts: 528
Joined: Sun Dec 09, 2018 4:51 pm

Re: Lacrosse Analytics

Post by Hawkeye »

laxreference wrote: Thu May 09, 2019 4:05 pm Let's assume that every team wins their opening round game. If you look at their projected title odds in that scenario and subtract their current odds, you get a list of the teams for whom a win would have the largest effect in terms of changing our perception of them. Even if Penn State wins their game, it's not really going to change our perception of their chances much. Yale vs Georgetown on the other hand: different story.

Here is the full post
Your model sure does like Georgetown a lot. Hmmmm.
Georgia Tech alumnus
2019 Georgia Tech lacrosse final record: 18-2; MCLA semifinalist
-
College lacrosse television schedules: https://markmhart9.wixsite.com/mysite
Post Reply

Return to “D1 MENS LACROSSE”